guild-agents 1.3.0 → 1.4.0

package/src/templates/skills/guild-specialize/evals/evals.json

@@ -0,0 +1,54 @@
+{
+  "skill": "guild-specialize",
+  "evals": [
+    {
+      "id": "gs-has-core-steps",
+      "description": "Guild specialize has read, explore, enrich, specialize, confirm, commit steps",
+      "expectations": [
+        { "text": "Has read-base step", "assertion": "step-exists:read-base" },
+        { "text": "Has explore-project step", "assertion": "step-exists:explore-project" },
+        { "text": "Has enrich-claude-md step", "assertion": "step-exists:enrich-claude-md" },
+        { "text": "Has specialize-agents step", "assertion": "step-exists:specialize-agents" },
+        { "text": "Has confirm step", "assertion": "step-exists:confirm" },
+        { "text": "Has commit-enrichment step", "assertion": "step-exists:commit-enrichment" }
+      ]
+    },
+    {
+      "id": "gs-enrichment-uses-reasoning",
+      "description": "CLAUDE.md enrichment uses reasoning tier (opus)",
+      "expectations": [
+        { "text": "enrich-claude-md uses reasoning", "assertion": "step-model-tier:enrich-claude-md:reasoning" }
+      ]
+    },
+    {
+      "id": "gs-agents-use-execution",
+      "description": "Agent specialization uses execution tier (sonnet)",
+      "expectations": [
+        { "text": "specialize-agents uses execution", "assertion": "step-model-tier:specialize-agents:execution" }
+      ]
+    },
+    {
+      "id": "gs-gates",
+      "description": "Gates at exploration and confirmation",
+      "expectations": [
+        { "text": "explore-project has gate", "assertion": "gate-exists:explore-project" },
+        { "text": "confirm has gate", "assertion": "gate-exists:confirm" }
+      ]
+    },
+    {
+      "id": "gs-tech-lead-role",
+      "description": "Enrichment and specialization use tech-lead role",
+      "expectations": [
+        { "text": "enrich-claude-md is tech-lead", "assertion": "step-role:enrich-claude-md:tech-lead" },
+        { "text": "specialize-agents is tech-lead", "assertion": "step-role:specialize-agents:tech-lead" }
+      ]
+    },
+    {
+      "id": "gs-minimum-steps",
+      "description": "Has at least 6 steps",
+      "expectations": [
+        { "text": "At least 6 steps", "assertion": "step-count:6" }
+      ]
+    }
+  ]
+}

package/src/templates/skills/guild-specialize/evals/triggers.json

@@ -0,0 +1,16 @@
+{
+  "skill": "guild-specialize",
+  "matcherType": "keyword",
+  "description": "Enriches CLAUDE.md by exploring the project and specializes agents to the real stack",
+  "threshold": 0.3,
+  "tests": [
+    { "prompt": "specialize the agents for this project", "shouldTrigger": true },
+    { "prompt": "enrich CLAUDE.md with the project stack", "shouldTrigger": true },
+    { "prompt": "explore the project and specialize agents", "shouldTrigger": true },
+    { "prompt": "set up Guild for this codebase", "shouldTrigger": true, "keywordExpected": false },
+    { "prompt": "create a pull request", "shouldTrigger": false },
+    { "prompt": "review my code", "shouldTrigger": false },
+    { "prompt": "debug this bug", "shouldTrigger": false },
+    { "prompt": "save my session", "shouldTrigger": false }
+  ]
+}

package/src/templates/skills/new-feature/evals/evals.json

@@ -0,0 +1,41 @@
+{
+  "skill": "new-feature",
+  "evals": [
+    {
+      "id": "nf-has-core-steps",
+      "description": "New feature has name, branch, session, confirm steps",
+      "expectations": [
+        { "text": "Has get-name step", "assertion": "step-exists:get-name" },
+        { "text": "Has create-branch step", "assertion": "step-exists:create-branch" },
+        { "text": "Has update-session step", "assertion": "step-exists:update-session" },
+        { "text": "Has confirm step", "assertion": "step-exists:confirm" }
+      ]
+    },
+    {
+      "id": "nf-optional-issue",
+      "description": "GitHub issue creation step exists",
+      "expectations": [
+        { "text": "Has create-issue step", "assertion": "step-exists:create-issue" }
+      ]
+    },
+    {
+      "id": "nf-gates",
+      "description": "Gates at name input, session update, and confirmation",
+      "expectations": [
+        { "text": "get-name has gate", "assertion": "gate-exists:get-name" },
+        { "text": "update-session has gate", "assertion": "gate-exists:update-session" },
+        { "text": "confirm has gate", "assertion": "gate-exists:confirm" }
+      ]
+    },
+    {
+      "id": "nf-all-system",
+      "description": "All steps are system role",
+      "expectations": [
+        { "text": "get-name is system", "assertion": "step-role:get-name:system" },
+        { "text": "create-branch is system", "assertion": "step-role:create-branch:system" },
+        { "text": "update-session is system", "assertion": "step-role:update-session:system" },
+        { "text": "confirm is system", "assertion": "step-role:confirm:system" }
+      ]
+    }
+  ]
+}

package/src/templates/skills/new-feature/evals/triggers.json

@@ -0,0 +1,16 @@
+{
+  "skill": "new-feature",
+  "matcherType": "keyword",
+  "description": "Creates branch and scaffold for a new feature",
+  "threshold": 0.3,
+  "tests": [
+    { "prompt": "create a new feature branch", "shouldTrigger": true },
+    { "prompt": "scaffold a new feature", "shouldTrigger": true },
+    { "prompt": "start a new feature called user-auth", "shouldTrigger": true },
+    { "prompt": "prepare the branch for a new feature", "shouldTrigger": true },
+    { "prompt": "review my code", "shouldTrigger": false },
+    { "prompt": "create a pull request", "shouldTrigger": false },
+    { "prompt": "debug this bug", "shouldTrigger": false },
+    { "prompt": "save my session", "shouldTrigger": false }
+  ]
+}

package/src/templates/skills/qa-cycle/evals/evals.json

@@ -0,0 +1,46 @@
+{
+  "skill": "qa-cycle",
+  "evals": [
+    {
+      "id": "qa-has-core-steps",
+      "description": "QA cycle has pre-gate, validate, bugfix, post-gate steps",
+      "expectations": [
+        { "text": "Has gate-pre-qa step", "assertion": "step-exists:gate-pre-qa" },
+        { "text": "Has qa-validate step", "assertion": "step-exists:qa-validate" },
+        { "text": "Has bugfix step", "assertion": "step-exists:bugfix" },
+        { "text": "Has gate-post-qa step", "assertion": "step-exists:gate-post-qa" }
+      ]
+    },
+    {
+      "id": "qa-agent-roles",
+      "description": "QA and bugfix steps use correct agent roles",
+      "expectations": [
+        { "text": "qa-validate uses qa role", "assertion": "step-role:qa-validate:qa" },
+        { "text": "bugfix uses bugfix role", "assertion": "step-role:bugfix:bugfix" }
+      ]
+    },
+    {
+      "id": "qa-execution-tier",
+      "description": "QA and bugfix use execution tier (sonnet)",
+      "expectations": [
+        { "text": "qa-validate uses execution", "assertion": "step-model-tier:qa-validate:execution" },
+        { "text": "bugfix uses execution", "assertion": "step-model-tier:bugfix:execution" }
+      ]
+    },
+    {
+      "id": "qa-gates",
+      "description": "Pre and post QA gates exist",
+      "expectations": [
+        { "text": "gate-pre-qa has gate", "assertion": "gate-exists:gate-pre-qa" },
+        { "text": "gate-post-qa has gate", "assertion": "gate-exists:gate-post-qa" }
+      ]
+    },
+    {
+      "id": "qa-bugfix-requires-report",
+      "description": "Bugfix step requires QA report",
+      "expectations": [
+        { "text": "bugfix requires qa-report", "assertion": "step-requires:bugfix:qa-report" }
+      ]
+    }
+  ]
+}

package/src/templates/skills/qa-cycle/evals/triggers.json

@@ -0,0 +1,16 @@
+{
+  "skill": "qa-cycle",
+  "matcherType": "keyword",
+  "description": "QA + bugfix cycle until it passes",
+  "threshold": 0.3,
+  "tests": [
+    { "prompt": "run a QA cycle on this", "shouldTrigger": true },
+    { "prompt": "QA and bugfix until it passes", "shouldTrigger": true },
+    { "prompt": "validate with QA cycle", "shouldTrigger": true },
+    { "prompt": "check if my implementation is correct", "shouldTrigger": true, "keywordExpected": false },
+    { "prompt": "create a pull request", "shouldTrigger": false },
+    { "prompt": "review my code", "shouldTrigger": false },
+    { "prompt": "save my session", "shouldTrigger": false },
+    { "prompt": "start a new feature", "shouldTrigger": false }
+  ]
+}

package/src/templates/skills/re-specialize/evals/evals.json

@@ -0,0 +1,48 @@
+{
+  "skill": "re-specialize",
+  "evals": [
+    {
+      "id": "rs-has-core-steps",
+      "description": "Re-specialize has read, explore, check-zones, regenerate, update, confirm, commit steps",
+      "expectations": [
+        { "text": "Has read-current step", "assertion": "step-exists:read-current" },
+        { "text": "Has explore-project step", "assertion": "step-exists:explore-project" },
+        { "text": "Has check-zones step", "assertion": "step-exists:check-zones" },
+        { "text": "Has regenerate-zones step", "assertion": "step-exists:regenerate-zones" },
+        { "text": "Has update-agents step", "assertion": "step-exists:update-agents" },
+        { "text": "Has confirm step", "assertion": "step-exists:confirm" },
+        { "text": "Has commit step", "assertion": "step-exists:commit" }
+      ]
+    },
+    {
+      "id": "rs-reasoning-for-regeneration",
+      "description": "Zone regeneration uses reasoning tier (opus)",
+      "expectations": [
+        { "text": "regenerate-zones uses reasoning", "assertion": "step-model-tier:regenerate-zones:reasoning" }
+      ]
+    },
+    {
+      "id": "rs-execution-for-agents",
+      "description": "Agent updates use execution tier (sonnet)",
+      "expectations": [
+        { "text": "update-agents uses execution", "assertion": "step-model-tier:update-agents:execution" }
+      ]
+    },
+    {
+      "id": "rs-gates",
+      "description": "Gates at exploration, zone check, and confirmation",
+      "expectations": [
+        { "text": "explore-project has gate", "assertion": "gate-exists:explore-project" },
+        { "text": "check-zones has gate", "assertion": "gate-exists:check-zones" },
+        { "text": "confirm has gate", "assertion": "gate-exists:confirm" }
+      ]
+    },
+    {
+      "id": "rs-minimum-steps",
+      "description": "Has at least 7 steps",
+      "expectations": [
+        { "text": "At least 7 steps", "assertion": "step-count:7" }
+      ]
+    }
+  ]
+}

package/src/templates/skills/re-specialize/evals/triggers.json

@@ -0,0 +1,16 @@
+{
+  "skill": "re-specialize",
+  "matcherType": "keyword",
+  "description": "Incremental re-specialization — re-scans the project and updates only auto-generated zones in CLAUDE.md and agents",
+  "threshold": 0.3,
+  "tests": [
+    { "prompt": "run incremental re-specialization", "shouldTrigger": true },
+    { "prompt": "re-scan the project and update CLAUDE.md zones", "shouldTrigger": true },
+    { "prompt": "update the auto-generated zones in agents", "shouldTrigger": true },
+    { "prompt": "my stack changed, update the agents", "shouldTrigger": true, "keywordExpected": false },
+    { "prompt": "create a pull request", "shouldTrigger": false },
+    { "prompt": "review my code", "shouldTrigger": false },
+    { "prompt": "debug this bug", "shouldTrigger": false },
+    { "prompt": "save my session", "shouldTrigger": false }
+  ]
+}

package/src/templates/skills/review/evals/evals.json

@@ -0,0 +1,43 @@
+{
+  "skill": "review",
+  "evals": [
+    {
+      "id": "rv-has-core-steps",
+      "description": "Review has gather-diff, review, and present steps",
+      "expectations": [
+        { "text": "Has gather-diff step", "assertion": "step-exists:gather-diff" },
+        { "text": "Has review step", "assertion": "step-exists:review" },
+        { "text": "Has present step", "assertion": "step-exists:present" }
+      ]
+    },
+    {
+      "id": "rv-reviewer-role",
+      "description": "Review step uses code-reviewer role",
+      "expectations": [
+        { "text": "review uses code-reviewer role", "assertion": "step-role:review:code-reviewer" }
+      ]
+    },
+    {
+      "id": "rv-reasoning-tier",
+      "description": "Code review uses reasoning tier (opus)",
+      "expectations": [
+        { "text": "review uses reasoning tier", "assertion": "step-model-tier:review:reasoning" }
+      ]
+    },
+    {
+      "id": "rv-presentation-gate",
+      "description": "Present step has gate for user review",
+      "expectations": [
+        { "text": "present has gate", "assertion": "gate-exists:present" }
+      ]
+    },
+    {
+      "id": "rv-dependencies",
+      "description": "Review requires diff content, present requires report",
+      "expectations": [
+        { "text": "review requires diff-content", "assertion": "step-requires:review:diff-content" },
+        { "text": "present requires review-report", "assertion": "step-requires:present:review-report" }
+      ]
+    }
+  ]
+}

package/src/templates/skills/review/evals/triggers.json

@@ -0,0 +1,16 @@
+{
+  "skill": "review",
+  "matcherType": "keyword",
+  "description": "Standalone code review on the current diff",
+  "threshold": 0.3,
+  "tests": [
+    { "prompt": "review my code", "shouldTrigger": true },
+    { "prompt": "do a code review on the current changes", "shouldTrigger": true },
+    { "prompt": "check my diff for issues", "shouldTrigger": true },
+    { "prompt": "review the current diff", "shouldTrigger": true },
+    { "prompt": "create a pull request", "shouldTrigger": false },
+    { "prompt": "save my session", "shouldTrigger": false },
+    { "prompt": "what phase am I in", "shouldTrigger": false },
+    { "prompt": "start a new feature", "shouldTrigger": false }
+  ]
+}

package/src/templates/skills/session-end/evals/evals.json

@@ -0,0 +1,40 @@
+{
+  "skill": "session-end",
+  "evals": [
+    {
+      "id": "se-has-core-steps",
+      "description": "Session end has gather, update, wip-commit, confirm steps",
+      "expectations": [
+        { "text": "Has gather-state step", "assertion": "step-exists:gather-state" },
+        { "text": "Has update-session step", "assertion": "step-exists:update-session" },
+        { "text": "Has commit-wip step", "assertion": "step-exists:commit-wip" },
+        { "text": "Has confirm step", "assertion": "step-exists:confirm" }
+      ]
+    },
+    {
+      "id": "se-all-system",
+      "description": "All steps are system role",
+      "expectations": [
+        { "text": "gather-state is system", "assertion": "step-role:gather-state:system" },
+        { "text": "update-session is system", "assertion": "step-role:update-session:system" },
+        { "text": "commit-wip is system", "assertion": "step-role:commit-wip:system" },
+        { "text": "confirm is system", "assertion": "step-role:confirm:system" }
+      ]
+    },
+    {
+      "id": "se-gates",
+      "description": "Gates at session update and confirmation",
+      "expectations": [
+        { "text": "update-session has gate", "assertion": "gate-exists:update-session" },
+        { "text": "confirm has gate", "assertion": "gate-exists:confirm" }
+      ]
+    },
+    {
+      "id": "se-wip-requires-files",
+      "description": "WIP commit requires knowledge of modified files",
+      "expectations": [
+        { "text": "commit-wip requires modified-files", "assertion": "step-requires:commit-wip:modified-files" }
+      ]
+    }
+  ]
+}

package/src/templates/skills/session-end/evals/triggers.json

@@ -0,0 +1,16 @@
+{
+  "skill": "session-end",
+  "matcherType": "keyword",
+  "description": "Saves current state to SESSION.md",
+  "threshold": 0.3,
+  "tests": [
+    { "prompt": "save the session state", "shouldTrigger": true },
+    { "prompt": "end my session and save to SESSION.md", "shouldTrigger": true },
+    { "prompt": "save current state to SESSION", "shouldTrigger": true },
+    { "prompt": "I'm done for today, save my progress", "shouldTrigger": true, "keywordExpected": false },
+    { "prompt": "start my session", "shouldTrigger": false },
+    { "prompt": "create a pull request", "shouldTrigger": false },
+    { "prompt": "review my code", "shouldTrigger": false },
+    { "prompt": "debug this bug", "shouldTrigger": false }
+  ]
+}

package/src/templates/skills/session-start/evals/evals.json

@@ -0,0 +1,50 @@
+{
+  "skill": "session-start",
+  "evals": [
+    {
+      "id": "ss-has-core-steps",
+      "description": "Session start has load, detect, present, suggest, update steps",
+      "expectations": [
+        { "text": "Has load-context step", "assertion": "step-exists:load-context" },
+        { "text": "Has detect-resumable step", "assertion": "step-exists:detect-resumable" },
+        { "text": "Has present-state step", "assertion": "step-exists:present-state" },
+        { "text": "Has suggest-continuation step", "assertion": "step-exists:suggest-continuation" },
+        { "text": "Has update-session step", "assertion": "step-exists:update-session" }
+      ]
+    },
+    {
+      "id": "ss-all-system",
+      "description": "All steps are system role",
+      "expectations": [
+        { "text": "load-context is system", "assertion": "step-role:load-context:system" },
+        { "text": "detect-resumable is system", "assertion": "step-role:detect-resumable:system" },
+        { "text": "present-state is system", "assertion": "step-role:present-state:system" },
+        { "text": "suggest-continuation is system", "assertion": "step-role:suggest-continuation:system" },
+        { "text": "update-session is system", "assertion": "step-role:update-session:system" }
+      ]
+    },
+    {
+      "id": "ss-gates",
+      "description": "Gates at presentation, suggestion, and session update",
+      "expectations": [
+        { "text": "present-state has gate", "assertion": "gate-exists:present-state" },
+        { "text": "suggest-continuation has gate", "assertion": "gate-exists:suggest-continuation" },
+        { "text": "update-session has gate", "assertion": "gate-exists:update-session" }
+      ]
+    },
+    {
+      "id": "ss-detect-requires-session",
+      "description": "Detect-resumable requires session state",
+      "expectations": [
+        { "text": "detect-resumable requires session-md", "assertion": "step-requires:detect-resumable:session-md" }
+      ]
+    },
+    {
+      "id": "ss-minimum-steps",
+      "description": "Has at least 5 steps",
+      "expectations": [
+        { "text": "At least 5 steps", "assertion": "step-count:5" }
+      ]
+    }
+  ]
+}

package/src/templates/skills/session-start/evals/triggers.json

@@ -0,0 +1,16 @@
+{
+  "skill": "session-start",
+  "matcherType": "keyword",
+  "description": "Loads context and resumes work from SESSION.md",
+  "threshold": 0.3,
+  "tests": [
+    { "prompt": "load my session context", "shouldTrigger": true },
+    { "prompt": "resume work from SESSION.md", "shouldTrigger": true },
+    { "prompt": "start session and load context", "shouldTrigger": true },
+    { "prompt": "where did I leave off", "shouldTrigger": true, "keywordExpected": false },
+    { "prompt": "save my progress", "shouldTrigger": false },
+    { "prompt": "create a new feature", "shouldTrigger": false },
+    { "prompt": "review my code", "shouldTrigger": false },
+    { "prompt": "run the tests", "shouldTrigger": false }
+  ]
+}

package/src/templates/skills/status/evals/evals.json

@@ -0,0 +1,40 @@
+{
+  "skill": "status",
+  "evals": [
+    {
+      "id": "st-has-core-steps",
+      "description": "Status has read-state, scan-resources, present steps",
+      "expectations": [
+        { "text": "Has read-state step", "assertion": "step-exists:read-state" },
+        { "text": "Has scan-resources step", "assertion": "step-exists:scan-resources" },
+        { "text": "Has present-status step", "assertion": "step-exists:present-status" }
+      ]
+    },
+    {
+      "id": "st-all-system",
+      "description": "All steps are system role",
+      "expectations": [
+        { "text": "read-state is system", "assertion": "step-role:read-state:system" },
+        { "text": "scan-resources is system", "assertion": "step-role:scan-resources:system" },
+        { "text": "present-status is system", "assertion": "step-role:present-status:system" }
+      ]
+    },
+    {
+      "id": "st-presentation-gate",
+      "description": "Present-status has gate",
+      "expectations": [
+        { "text": "present-status has gate", "assertion": "gate-exists:present-status" }
+      ]
+    },
+    {
+      "id": "st-dependencies",
+      "description": "Present-status requires project and session data",
+      "expectations": [
+        { "text": "present-status requires project-md", "assertion": "step-requires:present-status:project-md" },
+        { "text": "present-status requires session-md", "assertion": "step-requires:present-status:session-md" },
+        { "text": "present-status requires agent-list", "assertion": "step-requires:present-status:agent-list" },
+        { "text": "present-status requires skill-list", "assertion": "step-requires:present-status:skill-list" }
+      ]
+    }
+  ]
+}

package/src/templates/skills/status/evals/triggers.json

@@ -0,0 +1,16 @@
+{
+  "skill": "status",
+  "matcherType": "keyword",
+  "description": "Shows current project and session state",
+  "threshold": 0.3,
+  "tests": [
+    { "prompt": "show the project status", "shouldTrigger": true },
+    { "prompt": "show current session state", "shouldTrigger": true },
+    { "prompt": "what is the current project state", "shouldTrigger": true },
+    { "prompt": "how is the project going", "shouldTrigger": true },
+    { "prompt": "create a pull request", "shouldTrigger": false },
+    { "prompt": "review my code", "shouldTrigger": false },
+    { "prompt": "debug this bug", "shouldTrigger": false },
+    { "prompt": "save my session", "shouldTrigger": false }
+  ]
+}

package/src/templates/skills/tdd/evals/triggers.json

@@ -0,0 +1,16 @@
+{
+  "skill": "tdd",
+  "matcherType": "keyword",
+  "description": "Discipline skill — TDD red-green-refactor cycle. Use when implementing any feature or bugfix, before writing implementation code.",
+  "threshold": 0.3,
+  "tests": [
+    { "prompt": "use TDD for this implementation", "shouldTrigger": true },
+    { "prompt": "red green refactor cycle", "shouldTrigger": true },
+    { "prompt": "follow the TDD red-green-refactor discipline", "shouldTrigger": true },
+    { "prompt": "implement this feature with TDD", "shouldTrigger": true },
+    { "prompt": "create a pull request", "shouldTrigger": false },
+    { "prompt": "review my code", "shouldTrigger": false },
+    { "prompt": "save my session", "shouldTrigger": false },
+    { "prompt": "show project status", "shouldTrigger": false }
+  ]
+}

package/src/templates/skills/verify/evals/triggers.json

@@ -0,0 +1,16 @@
+{
+  "skill": "verify",
+  "matcherType": "keyword",
+  "description": "Discipline skill — verification before completion. Use when about to claim work is complete, fixed, or passing, before committing or creating PRs.",
+  "threshold": 0.3,
+  "tests": [
+    { "prompt": "verify before committing", "shouldTrigger": true },
+    { "prompt": "run verification before completion", "shouldTrigger": true },
+    { "prompt": "verify the work is complete", "shouldTrigger": true },
+    { "prompt": "make sure everything passes before the PR", "shouldTrigger": true, "keywordExpected": false },
+    { "prompt": "create a pull request", "shouldTrigger": false },
+    { "prompt": "review my code", "shouldTrigger": false },
+    { "prompt": "debug this bug", "shouldTrigger": false },
+    { "prompt": "save my session", "shouldTrigger": false }
+  ]
+}